A regional study of in-situ thermal conductivity of soil based on artificial neural network model

The in-situ thermal conductivity of the soil is an important parameter for designing a ground source heat pump system (GSHPs) with vertical boreholes, and this parameter is mainly obtained using in-situ thermal response tests (TRT). However, TRT requires the duration of more than 48 h and constant power during the heating process. If there is a power outage or malfunction during TRT, it is necessary to wait until ground temperature returns to the original value before re-testing, which is a long time and a large investment. To predict the in-situ thermal conductivity of soil accurately, this study develops an artificial neural network (ANN) model. Based on soil properties and groundwater characteristics of the test area, a new system of explanatory variables is proposed for predicting the in-situ thermal conductivity. A dataset of explanatory variables was proposed after in-situ TRT and investigations. The explanatory variables in dataset were proposed as stratigraphic type, weighted thermal conductivity of bedrock, aquifer thickness, permeability coefficient and groundwater depth. These five explanatory variables provide a comprehensive and detailed description of the borehole. The ANN model achieved the coefficient of determination R2 of 0.96815 and the average error of 6.3% between predicted and actual values in regions, which demonstrates it has good generalization ability. Therefore, this ANN model can be applied to obtain the in-situ thermal conductivity without massive in-situ TRT in similar regions. In addition, the contributions in ANN model of weighted thermal conductivity of bedrock, stratigraphic type, aquifer thickness, permeability coefficient and groundwater depth are 40.1%, 11.2%, 18.3%, 17.6% and 12.8% respectively. (C) 2021 Elsevier B.V. All rights reserved.

Automated summary

This study develops an artificial neural network (ANN) model to accurately predict the in-situ thermal conductivity of soil. By using a set of explanatory variables including stratigraphic type, weighted thermal conductivity of bedrock, aquifer thickness, permeability coefficient, and groundwater depth, the model is able to achieve accurate predictions in similar regions and demonstrates good generalization ability.